|
35 | 35 | #define ggml_quantize_mat_q8_0_4x8_generic ggml_quantize_mat_q8_0_4x8 |
36 | 36 | #define ggml_quantize_mat_q8_K_4x4_generic ggml_quantize_mat_q8_K_4x4 |
37 | 37 | #define ggml_quantize_mat_q8_K_4x8_generic ggml_quantize_mat_q8_K_4x8 |
| 38 | +#define ggml_repack_mat_f16_7x1_generic ggml_repack_mat_f16_7x1 |
| 39 | +#define ggml_repack_mat_f32_7x1_generic ggml_repack_mat_f32_7x1 |
38 | 40 | #define ggml_gemv_q4_0_4x4_q8_0_generic ggml_gemv_q4_0_4x4_q8_0 |
39 | 41 | #define ggml_gemv_q4_0_4x8_q8_0_generic ggml_gemv_q4_0_4x8_q8_0 |
40 | 42 | #define ggml_gemv_q4_0_8x8_q8_0_generic ggml_gemv_q4_0_8x8_q8_0 |
|
43 | 45 | #define ggml_gemv_q2_K_8x8_q8_K_generic ggml_gemv_q2_K_8x8_q8_K |
44 | 46 | #define ggml_gemv_iq4_nl_4x4_q8_0_generic ggml_gemv_iq4_nl_4x4_q8_0 |
45 | 47 | #define ggml_gemv_iq4_nl_8x8_q8_0_generic ggml_gemv_iq4_nl_8x8_q8_0 |
| 48 | +#define ggml_gemv_f16_1x16_f16_generic ggml_gemv_f16_1x16_f16 |
| 49 | +#define ggml_gemv_f16_1x32_f16_generic ggml_gemv_f16_1x32_f16 |
| 50 | +#define ggml_gemv_f16_1x64_f16_generic ggml_gemv_f16_1x64_f16 |
| 51 | +#define ggml_gemv_f16_1x128_f16_generic ggml_gemv_f16_1x128_f16 |
| 52 | +#define ggml_gemv_f32_1x16_f32_generic ggml_gemv_f32_1x16_f32 |
| 53 | +#define ggml_gemv_f32_1x32_f32_generic ggml_gemv_f32_1x32_f32 |
| 54 | +#define ggml_gemv_f32_1x64_f32_generic ggml_gemv_f32_1x64_f32 |
| 55 | +#define ggml_gemv_f32_1x128_f32_generic ggml_gemv_f32_1x128_f32 |
46 | 56 | #define ggml_gemm_q4_0_4x4_q8_0_generic ggml_gemm_q4_0_4x4_q8_0 |
47 | 57 | #define ggml_gemm_q4_0_4x8_q8_0_generic ggml_gemm_q4_0_4x8_q8_0 |
48 | 58 | #define ggml_gemm_q4_0_8x8_q8_0_generic ggml_gemm_q4_0_8x8_q8_0 |
|
51 | 61 | #define ggml_gemm_q2_K_8x8_q8_K_generic ggml_gemm_q2_K_8x8_q8_K |
52 | 62 | #define ggml_gemm_iq4_nl_4x4_q8_0_generic ggml_gemm_iq4_nl_4x4_q8_0 |
53 | 63 | #define ggml_gemm_iq4_nl_8x8_q8_0_generic ggml_gemm_iq4_nl_8x8_q8_0 |
| 64 | +#define ggml_gemm_f16_7x1x16_f16_generic ggml_gemm_f16_7x1x16_f16 |
| 65 | +#define ggml_gemm_f16_7x1x32_f16_generic ggml_gemm_f16_7x1x32_f16 |
| 66 | +#define ggml_gemm_f16_7x1x64_f16_generic ggml_gemm_f16_7x1x64_f16 |
| 67 | +#define ggml_gemm_f16_7x1x128_f16_generic ggml_gemm_f16_7x1x128_f16 |
| 68 | +#define ggml_gemm_f32_7x1x16_f32_generic ggml_gemm_f32_7x1x16_f32 |
| 69 | +#define ggml_gemm_f32_7x1x32_f32_generic ggml_gemm_f32_7x1x32_f32 |
| 70 | +#define ggml_gemm_f32_7x1x64_f32_generic ggml_gemm_f32_7x1x64_f32 |
| 71 | +#define ggml_gemm_f32_7x1x128_f32_generic ggml_gemm_f32_7x1x128_f32 |
54 | 72 | #elif defined(__aarch64__) || defined(__arm__) || defined(_M_ARM) || defined(_M_ARM64) |
55 | 73 | // repack.cpp |
56 | 74 | #define ggml_quantize_mat_q8_K_4x4_generic ggml_quantize_mat_q8_K_4x4 |
57 | 75 | #define ggml_quantize_mat_q8_K_4x8_generic ggml_quantize_mat_q8_K_4x8 |
| 76 | +#define ggml_repack_mat_f16_7x1_generic ggml_repack_mat_f16_7x1 |
| 77 | +#define ggml_repack_mat_f32_7x1_generic ggml_repack_mat_f32_7x1 |
58 | 78 | #define ggml_gemv_iq4_nl_8x8_q8_0_generic ggml_gemv_iq4_nl_8x8_q8_0 |
59 | 79 | #define ggml_gemv_q2_K_8x8_q8_K_generic ggml_gemv_q2_K_8x8_q8_K |
| 80 | +#define ggml_gemv_f16_1x16_f16_generic ggml_gemv_f16_1x16_f16 |
| 81 | +#define ggml_gemv_f16_1x32_f16_generic ggml_gemv_f16_1x32_f16 |
| 82 | +#define ggml_gemv_f16_1x64_f16_generic ggml_gemv_f16_1x64_f16 |
| 83 | +#define ggml_gemv_f16_1x128_f16_generic ggml_gemv_f16_1x128_f16 |
| 84 | +#define ggml_gemv_f32_1x16_f32_generic ggml_gemv_f32_1x16_f32 |
| 85 | +#define ggml_gemv_f32_1x32_f32_generic ggml_gemv_f32_1x32_f32 |
| 86 | +#define ggml_gemv_f32_1x64_f32_generic ggml_gemv_f32_1x64_f32 |
| 87 | +#define ggml_gemv_f32_1x128_f32_generic ggml_gemv_f32_1x128_f32 |
60 | 88 | #define ggml_gemm_iq4_nl_8x8_q8_0_generic ggml_gemm_iq4_nl_8x8_q8_0 |
61 | 89 | #define ggml_gemm_q2_K_8x8_q8_K_generic ggml_gemm_q2_K_8x8_q8_K |
| 90 | +#define ggml_gemm_f16_7x1x16_f16_generic ggml_gemm_f16_7x1x16_f16 |
| 91 | +#define ggml_gemm_f16_7x1x32_f16_generic ggml_gemm_f16_7x1x32_f16 |
| 92 | +#define ggml_gemm_f16_7x1x64_f16_generic ggml_gemm_f16_7x1x64_f16 |
| 93 | +#define ggml_gemm_f16_7x1x128_f16_generic ggml_gemm_f16_7x1x128_f16 |
| 94 | +#define ggml_gemm_f32_7x1x16_f32_generic ggml_gemm_f32_7x1x16_f32 |
| 95 | +#define ggml_gemm_f32_7x1x32_f32_generic ggml_gemm_f32_7x1x32_f32 |
| 96 | +#define ggml_gemm_f32_7x1x64_f32_generic ggml_gemm_f32_7x1x64_f32 |
| 97 | +#define ggml_gemm_f32_7x1x128_f32_generic ggml_gemm_f32_7x1x128_f32 |
62 | 98 | #elif defined(__x86_64__) || defined(__i386__) || defined(_M_IX86) || defined(_M_X64) |
63 | 99 | // repack.cpp |
64 | 100 | #define ggml_quantize_mat_q8_0_4x4_generic ggml_quantize_mat_q8_0_4x4 |
65 | 101 | #define ggml_quantize_mat_q8_K_4x4_generic ggml_quantize_mat_q8_K_4x4 |
| 102 | +#define ggml_repack_mat_f16_7x1_generic ggml_repack_mat_f16_7x1 |
| 103 | +#define ggml_repack_mat_f32_7x1_generic ggml_repack_mat_f32_7x1 |
66 | 104 | #define ggml_gemv_q4_0_4x4_q8_0_generic ggml_gemv_q4_0_4x4_q8_0 |
67 | 105 | #define ggml_gemv_q4_0_4x8_q8_0_generic ggml_gemv_q4_0_4x8_q8_0 |
68 | 106 | #define ggml_gemv_q4_K_8x4_q8_K_generic ggml_gemv_q4_K_8x4_q8_K |
69 | 107 | #define ggml_gemv_iq4_nl_4x4_q8_0_generic ggml_gemv_iq4_nl_4x4_q8_0 |
| 108 | +#define ggml_gemv_f16_1x16_f16_generic ggml_gemv_f16_1x16_f16 |
| 109 | +#define ggml_gemv_f16_1x32_f16_generic ggml_gemv_f16_1x32_f16 |
| 110 | +#define ggml_gemv_f16_1x64_f16_generic ggml_gemv_f16_1x64_f16 |
| 111 | +#define ggml_gemv_f16_1x128_f16_generic ggml_gemv_f16_1x128_f16 |
| 112 | +#define ggml_gemv_f32_1x16_f32_generic ggml_gemv_f32_1x16_f32 |
| 113 | +#define ggml_gemv_f32_1x32_f32_generic ggml_gemv_f32_1x32_f32 |
| 114 | +#define ggml_gemv_f32_1x64_f32_generic ggml_gemv_f32_1x64_f32 |
| 115 | +#define ggml_gemv_f32_1x128_f32_generic ggml_gemv_f32_1x128_f32 |
70 | 116 | #define ggml_gemm_q4_0_4x4_q8_0_generic ggml_gemm_q4_0_4x4_q8_0 |
71 | 117 | #define ggml_gemm_q4_0_4x8_q8_0_generic ggml_gemm_q4_0_4x8_q8_0 |
72 | 118 | #define ggml_gemm_q4_K_8x4_q8_K_generic ggml_gemm_q4_K_8x4_q8_K |
73 | 119 | #define ggml_gemm_iq4_nl_4x4_q8_0_generic ggml_gemm_iq4_nl_4x4_q8_0 |
| 120 | +#define ggml_gemm_f16_7x1x16_f16_generic ggml_gemm_f16_7x1x16_f16 |
| 121 | +#define ggml_gemm_f16_7x1x32_f16_generic ggml_gemm_f16_7x1x32_f16 |
| 122 | +#define ggml_gemm_f16_7x1x64_f16_generic ggml_gemm_f16_7x1x64_f16 |
| 123 | +#define ggml_gemm_f16_7x1x128_f16_generic ggml_gemm_f16_7x1x128_f16 |
| 124 | +#define ggml_gemm_f32_7x1x16_f32_generic ggml_gemm_f32_7x1x16_f32 |
| 125 | +#define ggml_gemm_f32_7x1x32_f32_generic ggml_gemm_f32_7x1x32_f32 |
| 126 | +#define ggml_gemm_f32_7x1x64_f32_generic ggml_gemm_f32_7x1x64_f32 |
| 127 | +#define ggml_gemm_f32_7x1x128_f32_generic ggml_gemm_f32_7x1x128_f32 |
74 | 128 | #elif defined(__POWERPC__) || defined(__powerpc__) |
75 | 129 | // ref: https://github.com/ggml-org/llama.cpp/pull/14146#issuecomment-2972561679 |
76 | 130 | // quants.c |
|
83 | 137 | #define ggml_quantize_mat_q8_0_4x8_generic ggml_quantize_mat_q8_0_4x8 |
84 | 138 | #define ggml_quantize_mat_q8_K_4x4_generic ggml_quantize_mat_q8_K_4x4 |
85 | 139 | #define ggml_quantize_mat_q8_K_4x8_generic ggml_quantize_mat_q8_K_4x8 |
| 140 | +#define ggml_repack_mat_f16_7x1_generic ggml_repack_mat_f16_7x1 |
| 141 | +#define ggml_repack_mat_f32_7x1_generic ggml_repack_mat_f32_7x1 |
86 | 142 | #define ggml_gemv_q4_0_4x4_q8_0_generic ggml_gemv_q4_0_4x4_q8_0 |
87 | 143 | #define ggml_gemv_q4_0_4x8_q8_0_generic ggml_gemv_q4_0_4x8_q8_0 |
88 | 144 | #define ggml_gemv_q4_0_8x8_q8_0_generic ggml_gemv_q4_0_8x8_q8_0 |
|
91 | 147 | #define ggml_gemv_q2_K_8x8_q8_K_generic ggml_gemv_q2_K_8x8_q8_K |
92 | 148 | #define ggml_gemv_iq4_nl_4x4_q8_0_generic ggml_gemv_iq4_nl_4x4_q8_0 |
93 | 149 | #define ggml_gemv_iq4_nl_8x8_q8_0_generic ggml_gemv_iq4_nl_8x8_q8_0 |
| 150 | +#define ggml_gemv_f16_1x16_f16_generic ggml_gemv_f16_1x16_f16 |
| 151 | +#define ggml_gemv_f16_1x32_f16_generic ggml_gemv_f16_1x32_f16 |
| 152 | +#define ggml_gemv_f16_1x64_f16_generic ggml_gemv_f16_1x64_f16 |
| 153 | +#define ggml_gemv_f16_1x128_f16_generic ggml_gemv_f16_1x128_f16 |
| 154 | +#define ggml_gemv_f32_1x16_f32_generic ggml_gemv_f32_1x16_f32 |
| 155 | +#define ggml_gemv_f32_1x32_f32_generic ggml_gemv_f32_1x32_f32 |
| 156 | +#define ggml_gemv_f32_1x64_f32_generic ggml_gemv_f32_1x64_f32 |
| 157 | +#define ggml_gemv_f32_1x128_f32_generic ggml_gemv_f32_1x128_f32 |
94 | 158 | #define ggml_gemm_q4_0_4x4_q8_0_generic ggml_gemm_q4_0_4x4_q8_0 |
95 | 159 | #define ggml_gemm_q4_0_4x8_q8_0_generic ggml_gemm_q4_0_4x8_q8_0 |
96 | 160 | #define ggml_gemm_q4_0_8x8_q8_0_generic ggml_gemm_q4_0_8x8_q8_0 |
|
99 | 163 | #define ggml_gemm_q2_K_8x8_q8_K_generic ggml_gemm_q2_K_8x8_q8_K |
100 | 164 | #define ggml_gemm_iq4_nl_4x4_q8_0_generic ggml_gemm_iq4_nl_4x4_q8_0 |
101 | 165 | #define ggml_gemm_iq4_nl_8x8_q8_0_generic ggml_gemm_iq4_nl_8x8_q8_0 |
| 166 | +#define ggml_gemm_f16_7x1x16_f16_generic ggml_gemm_f16_7x1x16_f16 |
| 167 | +#define ggml_gemm_f16_7x1x32_f16_generic ggml_gemm_f16_7x1x32_f16 |
| 168 | +#define ggml_gemm_f16_7x1x64_f16_generic ggml_gemm_f16_7x1x64_f16 |
| 169 | +#define ggml_gemm_f16_7x1x128_f16_generic ggml_gemm_f16_7x1x128_f16 |
| 170 | +#define ggml_gemm_f32_7x1x16_f32_generic ggml_gemm_f32_7x1x16_f32 |
| 171 | +#define ggml_gemm_f32_7x1x32_f32_generic ggml_gemm_f32_7x1x32_f32 |
| 172 | +#define ggml_gemm_f32_7x1x64_f32_generic ggml_gemm_f32_7x1x64_f32 |
| 173 | +#define ggml_gemm_f32_7x1x128_f32_generic ggml_gemm_f32_7x1x128_f32 |
102 | 174 | #elif defined(__loongarch64) |
103 | 175 | // quants.c |
104 | 176 | #define quantize_row_q8_K_generic quantize_row_q8_K |
|
111 | 183 | #define ggml_quantize_mat_q8_0_4x8_generic ggml_quantize_mat_q8_0_4x8 |
112 | 184 | #define ggml_quantize_mat_q8_K_4x4_generic ggml_quantize_mat_q8_K_4x4 |
113 | 185 | #define ggml_quantize_mat_q8_K_4x8_generic ggml_quantize_mat_q8_K_4x8 |
| 186 | +#define ggml_repack_mat_f16_7x1_generic ggml_repack_mat_f16_7x1 |
| 187 | +#define ggml_repack_mat_f32_7x1_generic ggml_repack_mat_f32_7x1 |
114 | 188 | #define ggml_gemv_q4_0_4x4_q8_0_generic ggml_gemv_q4_0_4x4_q8_0 |
115 | 189 | #define ggml_gemv_q4_0_4x8_q8_0_generic ggml_gemv_q4_0_4x8_q8_0 |
116 | 190 | #define ggml_gemv_q4_0_8x8_q8_0_generic ggml_gemv_q4_0_8x8_q8_0 |
|
119 | 193 | #define ggml_gemv_q2_K_8x8_q8_K_generic ggml_gemv_q2_K_8x8_q8_K |
120 | 194 | #define ggml_gemv_iq4_nl_4x4_q8_0_generic ggml_gemv_iq4_nl_4x4_q8_0 |
121 | 195 | #define ggml_gemv_iq4_nl_8x8_q8_0_generic ggml_gemv_iq4_nl_8x8_q8_0 |
| 196 | +#define ggml_gemv_f16_1x16_f16_generic ggml_gemv_f16_1x16_f16 |
| 197 | +#define ggml_gemv_f16_1x32_f16_generic ggml_gemv_f16_1x32_f16 |
| 198 | +#define ggml_gemv_f16_1x64_f16_generic ggml_gemv_f16_1x64_f16 |
| 199 | +#define ggml_gemv_f16_1x128_f16_generic ggml_gemv_f16_1x128_f16 |
| 200 | +#define ggml_gemv_f32_1x16_f32_generic ggml_gemv_f32_1x16_f32 |
| 201 | +#define ggml_gemv_f32_1x32_f32_generic ggml_gemv_f32_1x32_f32 |
| 202 | +#define ggml_gemv_f32_1x64_f32_generic ggml_gemv_f32_1x64_f32 |
| 203 | +#define ggml_gemv_f32_1x128_f32_generic ggml_gemv_f32_1x128_f32 |
122 | 204 | #define ggml_gemm_q4_0_4x4_q8_0_generic ggml_gemm_q4_0_4x4_q8_0 |
123 | 205 | #define ggml_gemm_q4_0_4x8_q8_0_generic ggml_gemm_q4_0_4x8_q8_0 |
124 | 206 | #define ggml_gemm_q4_0_8x8_q8_0_generic ggml_gemm_q4_0_8x8_q8_0 |
|
127 | 209 | #define ggml_gemm_q2_K_8x8_q8_K_generic ggml_gemm_q2_K_8x8_q8_K |
128 | 210 | #define ggml_gemm_iq4_nl_4x4_q8_0_generic ggml_gemm_iq4_nl_4x4_q8_0 |
129 | 211 | #define ggml_gemm_iq4_nl_8x8_q8_0_generic ggml_gemm_iq4_nl_8x8_q8_0 |
| 212 | +#define ggml_gemm_f16_7x1x16_f16_generic ggml_gemm_f16_7x1x16_f16 |
| 213 | +#define ggml_gemm_f16_7x1x32_f16_generic ggml_gemm_f16_7x1x32_f16 |
| 214 | +#define ggml_gemm_f16_7x1x64_f16_generic ggml_gemm_f16_7x1x64_f16 |
| 215 | +#define ggml_gemm_f16_7x1x128_f16_generic ggml_gemm_f16_7x1x128_f16 |
| 216 | +#define ggml_gemm_f32_7x1x16_f32_generic ggml_gemm_f32_7x1x16_f32 |
| 217 | +#define ggml_gemm_f32_7x1x32_f32_generic ggml_gemm_f32_7x1x32_f32 |
| 218 | +#define ggml_gemm_f32_7x1x64_f32_generic ggml_gemm_f32_7x1x64_f32 |
| 219 | +#define ggml_gemm_f32_7x1x128_f32_generic ggml_gemm_f32_7x1x128_f32 |
130 | 220 | #elif defined(__riscv) |
131 | 221 | // quants.c |
132 | 222 | #define quantize_row_q8_K_generic quantize_row_q8_K |
|
147 | 237 | #define ggml_quantize_mat_q8_0_4x8_generic ggml_quantize_mat_q8_0_4x8 |
148 | 238 | #define ggml_quantize_mat_q8_K_4x4_generic ggml_quantize_mat_q8_K_4x4 |
149 | 239 | #define ggml_quantize_mat_q8_K_4x8_generic ggml_quantize_mat_q8_K_4x8 |
| 240 | +#define ggml_repack_mat_f16_7x1_generic ggml_repack_mat_f16_7x1 |
| 241 | +#define ggml_repack_mat_f32_7x1_generic ggml_repack_mat_f32_7x1 |
150 | 242 | #define ggml_gemv_q4_0_4x4_q8_0_generic ggml_gemv_q4_0_4x4_q8_0 |
151 | 243 | #define ggml_gemv_q4_0_4x8_q8_0_generic ggml_gemv_q4_0_4x8_q8_0 |
152 | 244 | #define ggml_gemv_q4_K_8x4_q8_K_generic ggml_gemv_q4_K_8x4_q8_K |
|
179 | 271 | #define ggml_quantize_mat_q8_0_4x8_generic ggml_quantize_mat_q8_0_4x8 |
180 | 272 | #define ggml_quantize_mat_q8_K_4x4_generic ggml_quantize_mat_q8_K_4x4 |
181 | 273 | #define ggml_quantize_mat_q8_K_4x8_generic ggml_quantize_mat_q8_K_4x8 |
| 274 | +#define ggml_repack_mat_f16_7x1_generic ggml_repack_mat_f16_7x1 |
| 275 | +#define ggml_repack_mat_f32_7x1_generic ggml_repack_mat_f32_7x1 |
182 | 276 | #define ggml_gemv_q4_0_4x4_q8_0_generic ggml_gemv_q4_0_4x4_q8_0 |
183 | 277 | #define ggml_gemv_q4_0_4x8_q8_0_generic ggml_gemv_q4_0_4x8_q8_0 |
184 | 278 | #define ggml_gemv_q4_0_8x8_q8_0_generic ggml_gemv_q4_0_8x8_q8_0 |
|
187 | 281 | #define ggml_gemv_q2_K_8x8_q8_K_generic ggml_gemv_q2_K_8x8_q8_K |
188 | 282 | #define ggml_gemv_iq4_nl_4x4_q8_0_generic ggml_gemv_iq4_nl_4x4_q8_0 |
189 | 283 | #define ggml_gemv_iq4_nl_8x8_q8_0_generic ggml_gemv_iq4_nl_8x8_q8_0 |
| 284 | +#define ggml_gemv_f16_1x16_f16_generic ggml_gemv_f16_1x16_f16 |
| 285 | +#define ggml_gemv_f16_1x32_f16_generic ggml_gemv_f16_1x32_f16 |
| 286 | +#define ggml_gemv_f16_1x64_f16_generic ggml_gemv_f16_1x64_f16 |
| 287 | +#define ggml_gemv_f16_1x128_f16_generic ggml_gemv_f16_1x128_f16 |
| 288 | +#define ggml_gemv_f32_1x16_f32_generic ggml_gemv_f32_1x16_f32 |
| 289 | +#define ggml_gemv_f32_1x32_f32_generic ggml_gemv_f32_1x32_f32 |
| 290 | +#define ggml_gemv_f32_1x64_f32_generic ggml_gemv_f32_1x64_f32 |
| 291 | +#define ggml_gemv_f32_1x128_f32_generic ggml_gemv_f32_1x128_f32 |
190 | 292 | #define ggml_gemm_q4_0_4x4_q8_0_generic ggml_gemm_q4_0_4x4_q8_0 |
191 | 293 | #define ggml_gemm_q4_0_4x8_q8_0_generic ggml_gemm_q4_0_4x8_q8_0 |
192 | 294 | #define ggml_gemm_q4_0_8x8_q8_0_generic ggml_gemm_q4_0_8x8_q8_0 |
|
195 | 297 | #define ggml_gemm_q2_K_8x8_q8_K_generic ggml_gemm_q2_K_8x8_q8_K |
196 | 298 | #define ggml_gemm_iq4_nl_4x4_q8_0_generic ggml_gemm_iq4_nl_4x4_q8_0 |
197 | 299 | #define ggml_gemm_iq4_nl_8x8_q8_0_generic ggml_gemm_iq4_nl_8x8_q8_0 |
| 300 | +#define ggml_gemm_f16_7x1x16_f16_generic ggml_gemm_f16_7x1x16_f16 |
| 301 | +#define ggml_gemm_f16_7x1x32_f16_generic ggml_gemm_f16_7x1x32_f16 |
| 302 | +#define ggml_gemm_f16_7x1x64_f16_generic ggml_gemm_f16_7x1x64_f16 |
| 303 | +#define ggml_gemm_f16_7x1x128_f16_generic ggml_gemm_f16_7x1x128_f16 |
| 304 | +#define ggml_gemm_f32_7x1x16_f32_generic ggml_gemm_f32_7x1x16_f32 |
| 305 | +#define ggml_gemm_f32_7x1x32_f32_generic ggml_gemm_f32_7x1x32_f32 |
| 306 | +#define ggml_gemm_f32_7x1x64_f32_generic ggml_gemm_f32_7x1x64_f32 |
| 307 | +#define ggml_gemm_f32_7x1x128_f32_generic ggml_gemm_f32_7x1x128_f32 |
198 | 308 | #elif defined(__wasm__) |
199 | 309 | // quants.c |
200 | 310 | #define ggml_vec_dot_q4_1_q8_1_generic ggml_vec_dot_q4_1_q8_1 |
|
215 | 325 | #define ggml_quantize_mat_q8_0_4x8_generic ggml_quantize_mat_q8_0_4x8 |
216 | 326 | #define ggml_quantize_mat_q8_K_4x4_generic ggml_quantize_mat_q8_K_4x4 |
217 | 327 | #define ggml_quantize_mat_q8_K_4x8_generic ggml_quantize_mat_q8_K_4x8 |
| 328 | +#define ggml_repack_mat_f16_7x1_generic ggml_repack_mat_f16_7x1 |
| 329 | +#define ggml_repack_mat_f32_7x1_generic ggml_repack_mat_f32_7x1 |
218 | 330 | #define ggml_gemv_q4_0_4x4_q8_0_generic ggml_gemv_q4_0_4x4_q8_0 |
219 | 331 | #define ggml_gemv_q4_0_4x8_q8_0_generic ggml_gemv_q4_0_4x8_q8_0 |
220 | 332 | #define ggml_gemv_q4_0_8x8_q8_0_generic ggml_gemv_q4_0_8x8_q8_0 |
|
223 | 335 | #define ggml_gemv_q2_K_8x8_q8_K_generic ggml_gemv_q2_K_8x8_q8_K |
224 | 336 | #define ggml_gemv_iq4_nl_4x4_q8_0_generic ggml_gemv_iq4_nl_4x4_q8_0 |
225 | 337 | #define ggml_gemv_iq4_nl_8x8_q8_0_generic ggml_gemv_iq4_nl_8x8_q8_0 |
| 338 | +#define ggml_gemv_f16_1x16_f16_generic ggml_gemv_f16_1x16_f16 |
| 339 | +#define ggml_gemv_f16_1x32_f16_generic ggml_gemv_f16_1x32_f16 |
| 340 | +#define ggml_gemv_f16_1x64_f16_generic ggml_gemv_f16_1x64_f16 |
| 341 | +#define ggml_gemv_f16_1x128_f16_generic ggml_gemv_f16_1x128_f16 |
| 342 | +#define ggml_gemv_f32_1x16_f32_generic ggml_gemv_f32_1x16_f32 |
| 343 | +#define ggml_gemv_f32_1x32_f32_generic ggml_gemv_f32_1x32_f32 |
| 344 | +#define ggml_gemv_f32_1x64_f32_generic ggml_gemv_f32_1x64_f32 |
| 345 | +#define ggml_gemv_f32_1x128_f32_generic ggml_gemv_f32_1x128_f32 |
226 | 346 | #define ggml_gemm_q4_0_4x4_q8_0_generic ggml_gemm_q4_0_4x4_q8_0 |
227 | 347 | #define ggml_gemm_q4_0_4x8_q8_0_generic ggml_gemm_q4_0_4x8_q8_0 |
228 | 348 | #define ggml_gemm_q4_0_8x8_q8_0_generic ggml_gemm_q4_0_8x8_q8_0 |
|
231 | 351 | #define ggml_gemm_q2_K_8x8_q8_K_generic ggml_gemm_q2_K_8x8_q8_K |
232 | 352 | #define ggml_gemm_iq4_nl_4x4_q8_0_generic ggml_gemm_iq4_nl_4x4_q8_0 |
233 | 353 | #define ggml_gemm_iq4_nl_8x8_q8_0_generic ggml_gemm_iq4_nl_8x8_q8_0 |
| 354 | +#define ggml_gemm_f16_7x1x16_f16_generic ggml_gemm_f16_7x1x16_f16 |
| 355 | +#define ggml_gemm_f16_7x1x32_f16_generic ggml_gemm_f16_7x1x32_f16 |
| 356 | +#define ggml_gemm_f16_7x1x64_f16_generic ggml_gemm_f16_7x1x64_f16 |
| 357 | +#define ggml_gemm_f16_7x1x128_f16_generic ggml_gemm_f16_7x1x128_f16 |
| 358 | +#define ggml_gemm_f32_7x1x16_f32_generic ggml_gemm_f32_7x1x16_f32 |
| 359 | +#define ggml_gemm_f32_7x1x32_f32_generic ggml_gemm_f32_7x1x32_f32 |
| 360 | +#define ggml_gemm_f32_7x1x64_f32_generic ggml_gemm_f32_7x1x64_f32 |
| 361 | +#define ggml_gemm_f32_7x1x128_f32_generic ggml_gemm_f32_7x1x128_f32 |
234 | 362 | #endif |
0 commit comments